JetBrains Unveils Mellum: The Revolutionary Open AI Coding Model for Developers
JetBrains has recently launched its first open AI model for coding, making significant strides in the realm of software development tools. The new model, named Mellum, is designed to enhance coding efficiency and is now available on the popular AI development platform, Hugging Face.
Introducing Mellum: The Code-Generating AI Model
On Wednesday, JetBrains released Mellum, a sophisticated code-generating model that the company initially introduced within its software development suites last year. This advanced model has been trained on over 4 trillion tokens and boasts an impressive 4 billion parameters, specifically optimized for code completion tasks. This means it can effectively complete code snippets based on their contextual surroundings.
Understanding Tokens and Parameters
In the context of AI, parameters indicate a model’s problem-solving capabilities, while tokens represent the raw data the model processes. For perspective, one million tokens equate to approximately 30,000 lines of code.
Key Features of Mellum
According to JetBrains, Mellum was designed with several applications in mind, including:
- Integration into professional developer tools for intelligent code suggestions.
- AI-powered coding assistants to enhance productivity.
- Research on code understanding and generation.
- Educational applications for learning coding.
- Fine-tuning experiments for advanced users.
Mellum, which is licensed under Apache 2.0, was trained using a variety of datasets, including permissively licensed code from GitHub and articles from English-language Wikipedia. The training process lasted around 20 days, utilizing a cluster of 256 H200 Nvidia GPUs.
Getting Started with Mellum
While Mellum offers remarkable capabilities, it requires some initial setup. The base model is not ready for immediate use; users must fine-tune it first. JetBrains has made available a few pre-fine-tuned Mellum models specifically for Python development, although these are intended for exploring capabilities rather than for production deployment.
Addressing Security Challenges in AI-Generated Code
As AI-generated code transforms software development, it also raises important security concerns. A recent survey by Snyk revealed that over 50% of organizations encounter security issues with AI-produced code either occasionally or frequently.
JetBrains acknowledges that Mellum may inherit biases from the public codebases it was trained on, potentially resulting in code that mirrors styles prevalent in open-source repositories. Consequently, the code suggestions generated by Mellum may not always be secure or vulnerable-free.
The Future of Mellum
In a recent blog post, JetBrains expressed excitement for the future of Mellum, stating, “This is just the beginning. We’re not chasing generality — we’re building focus. If Mellum sparks even one meaningful experiment, contribution, or collaboration, we would consider it a win.”
For more information on JetBrains’ innovative tools and AI developments, visit their official website at JetBrains.
